A Stanford Medicine pilot study has found that an AI-powered tool can help doctors prepare hospital discharge summaries while easing cognitive burden and reducing reported burnout.
The in-house system, known as MedAgentBrief, was designed to condense complex patient histories into draft discharge summaries for physician review. Discharge summaries are essential for continuity of care, but can be time-consuming because doctors must summarise days or weeks of clinical information for outpatient providers.
The tool was piloted for 10 weeks at a Stanford Health Care unit at Sequoia Hospital. Eleven physicians received secure daily emails containing AI-generated summaries for each of their patients, using a format based on a best-practice discharge summary template.
Doctors were asked to review the summaries and provide feedback on inaccuracies, omissions, hallucinations and potential harm. Feedback on 100 AI-enabled summaries found omissions in 25% and inaccuracies in 20%, while hallucinations were rare at 2%.
Physicians rated most unedited summaries as having no harm potential, while some were judged to have mild harm potential. One summary was initially rated as likely to cause moderate harm, but independent reviewers later determined that it posed no risk. No severe harm was reported.
The pilot did not show large time savings. Physicians felt they saved more than 10 minutes per discharge summary, but electronic medical record logs suggested the actual savings were modest, at around three minutes at best. Researchers said the benefit may have come from reducing the effort required to write summaries from scratch.
Participating physicians also reported lower burnout scores. Researchers said the findings suggest AI-generated discharge summaries can support clinical documentation when used under physician supervision, while further evaluation will be needed as health systems consider vendor-based tools.
Why does it matter?
The pilot is useful because it focuses on a practical, high-friction healthcare workflow rather than a speculative AI use case. Discharge summaries affect continuity of care, and poor documentation can create risks for patients after they leave the hospital. Stanford's findings suggest AI may help reduce doctors' cognitive load, but also show why human review remains essential: omissions and inaccuracies were common enough that the tool should be treated as clinical support, not automation.